FaddeevPopov ghosts

Physics 230A, Spring 2007, Hitoshi Murayama

In this lecture notes, I describe what I discussed in class about Faddeev
Popov ghosts except for the material already described in PeskinSchroeder.

Why FaddeevPopov Determinant?

Suppose you have an integral whose integrand is highly symmetric. It is a

kind of waste to integrate over the entire volume when you know you are
adding the same integrand over and over again. You then would want to
reduce the integration volume using the symmetry of the integrand. This is
where the FaddeevPopov method comes in.
Let us look at this simple example. You have an integral of a rotationally
symmetric function f (r, ) which depends only on the radius r = |~x| and
other physical parameters . You start with a two-dimensional integral
Z
Z() = d2 xf (r, ).
(1)
In the end, we are only interested in the dependence of the result, not
the overall normalization. In this case, it is obvious what we should do. We
change the integration variables to the polar coordinates and throw away the
angular integral,
Z
Z 2
Z
Z
Z() =
rdr
df (r, ) = 2
rdrf (r, )
rdrf (r, ). (2)
0

Now we do the same exercise in a different way. In many cases, it is

not easy to find the change of variables to eliminate redundant variables
explicitly. It would be nice to have a method that would apply more generally
even when you dont know what the good variables are.
If I am naive, I may do the following which leads to a wrong result. I
argue that the integrand depends on the radius alone, and it shouldnt matter
along what angle I integrate. Then I can decide to integrate only along the
x-axis and still get the same answer up to an overall constant. I would write
Z
Z()
dxf (x, ).
(3)
0

We know this is wrong. Where did I make a mistake?

When I restricted the integral to the x-axis alone, I basically inserted (y)
into the integrand. The problem is that this delta function does not behave
nicely under rotation. To see this, let us check what happens when I rotate
the coordinates from y to y = y cos x sin :
y
1
.
(y) (y cos x sin ) = ( tan1 )
x |y sin x cos |

(4)

Namely it does not fix the angle without an additional dependent factor.
What we want is to just fix the angular integral, namely (), not (y).
We know this because we can separate variables easily in this simple case.
In more complicated cases, we dont necessarily know. What we can do is to
start with (y), but then compensate for the rotation we did above as:
Z
y
1=
d( tan1 )
x
0


Z
Z
 y 
 . (5)
=
d(y cos x sin )|y sin x cos | =
d(y ) 

0
0
We insert this expression into the original integral, and we find


Z
Z
 y 
2
 f (r, ).
d(y ) 
Z() = d x

0

(6)

We can now change the integration variable from y to y so that it becomes



Z
Z
 y 
2

 f (r, ).
Z() = d x
d(y) 
(7)


0
=0
Now there is no dependence in the integrand anymore, so that it just gives
a factor of . We also know


 y 


= |y sin x cos |=0 = |x|.
(8)
 
=0
Therefore,
Z
Z() =

d x(y)|x|f (r, ) =

Z
dx|x|f (r, ) = 2

dxxf (x, ).
0

(9)
2

This is exactly what we want. FaddeevPopov method is nothing but the

generalization of what we have done here.
Let us generalize the above example to D-dimensional integral. For
a rotationally invariant integrand f (r, ), we integrate over the entire Ddimensional volume
Z
Z() = dD xf (r, ).
(10)
We would like to impose the gauge fixing condition x2 = x3 = = xD = 0
and keep only x1 as the integration variable. There are D 1 rotations that
change the gauge fixing condition on the (1, i) plane,
x1 x1 cos 1i xi sin 1i ,
xi x1 sin 1i + xi cos 1i .

(11)
(12)

Following the FaddeevPopov method, we insert

 i 
Z
Z
 dx 
1
i
1
i
i
1 = d1i (x sin 1i +x cos 1i )|x cos 1i x sin 1i | = d1i (x1i )  1i  ,
d1i
(13)
for each i = 2, , D. Then the original integral becomes
Z
Z() =

d xf (r, )

D
Y
i=2

 i 
 dx1i 
.
d1i 

d1i (xi1i ) 

(14)

Rotate each of xi1i back to xi , we find

Z
Z() =

d xf (r, )

D
Y

 
d1i (x ) x1 
i

dx1 f (r, )|x1 |D1 .

(15)

i=2

We know this is the correct result because we could have gone to spherical
coordinates from the beginning. The final form still double counts the
integrand because x1 is integrated from to , but it is much more
efficient than the original full volume integral. Note that we didnt have
to think hard about the integration regions for the angles 1i because we
were not concerned with the overall normalization. Namely that we only
needed information about infinitesimal symmetry transformations, which is
the beauty of the FaddeevPopov method.
More to come.
3